use different xtts voice

src/auracast_translator/translator_server/translator_server.py

@@ -154,7 +154,10 @@ async def make_announcement(text: str, ep_group: EndpointGroup):
             ep_group.sampling_rate_hz,
             trans_conf.tts_system,
             trans_conf.tts_model,
-            return_lc3=True
+            return_lc3=True,
+            language=trans_conf.xtts_language,
+            speaker=trans_conf.xtts_speaker,
+            speaker_wav=trans_conf.xtts_speaker_wav
         )
         synthesis_tasks.append(task)
     

src/translator_models/translator_models.py

@@ -37,15 +37,16 @@ class TranslatorLangConfig(BaseModel):
     
     tts_system: str = 'piper'  # Options: 'piper', 'xtts'
     tts_model: str = 'de_DE-kerstin-low'  # For piper: model name, for xtts: unused
-    tts_language: str = 'de'  # Language code for XTTS
-    tts_speaker: Optional[str] = None  # Speaker name for XTTS
+    xtts_language: str = 'de'  # Language code for XTTS
+    xtts_speaker: Optional[str] = None  # Speaker name for XTTS
+    xtts_speaker_wav: Optional[str] = None  # Path to speaker sample for XTTS voice cloning
 
 
 class TranslatorConfig(BaseModel):
     deu: TranslatorLangConfig = TranslatorLangConfig(
         tts_system='xtts',
-        tts_language='de',
-        tts_speaker='Annmarie Nele'
+        xtts_language='de',
+        xtts_speaker_wav='female.wav'
     )
     eng: TranslatorLangConfig = TranslatorLangConfig(tts_model = 'en_GB-alba-medium')
     fra: TranslatorLangConfig = TranslatorLangConfig(tts_model = 'fr_FR-siwis-medium')

src/voice_provider/text_to_speech.py

@@ -103,6 +103,8 @@ def synthesize(
         
         # Generate audio samples using tts.tts
         if speaker_wav:
+            # expand path to speaker_wav folder
+            speaker_wav = os.path.join(os.path.dirname(__file__), 'speaker_wav', speaker_wav)
             log.info(f"Generating XTTS audio with speaker_wav: {speaker_wav}")
             audio_list = XTTS_MODEL.tts(text=text, speaker_wav=speaker_wav, language=language)
         else:
@@ -117,7 +119,7 @@ def synthesize(
                 f"min: {audio_np.min():.4f}, max: {audio_np.max():.4f}")
         
         # Resample from 24kHz to target sample rate with speedup factor
-        audio = resample_array(audio_np, xtts_sample_rate, target_sample_rate, speedup=1.05)
+        audio = resample_array(audio_np, xtts_sample_rate, target_sample_rate)
         
         log.info(f"XTTS synthesis completed in {time.time() - start:.2f} seconds")
 
@@ -196,8 +198,8 @@ if __name__ == '__main__':
     sf.write('hello_piper.wav', audio_piper, target_rate)
 
     # Demo of XTTS with Annmarie Nele for German
-    german_speaker = 'Annmarie Nele'
-    print(f"Testing XTTS with German language using speaker: {german_speaker}")
+    speaker_wav = 'female.wav'
+    print(f"Testing XTTS with German language using speaker: {speaker_wav}")
     text_to_synthesize = "Dies ist ein Test der XTTS Stimme auf Deutsch mit Annmarie Nele als Sprecherin."
     
     audio_xtts = synthesize(
@@ -205,22 +207,11 @@ if __name__ == '__main__':
         target_sample_rate=target_rate,
         framework='xtts',
         language='de',
-        speaker=german_speaker,
+        speaker_wav=speaker_wav,
         return_lc3=False
     )
     
     # Save the wav file
     sf.write('hello_xtts_german.wav', audio_xtts, target_rate)
     
-    # Also test with LC3 encoding
-    lc3_xtts = synthesize(
-        text=text_to_synthesize,
-        target_sample_rate=target_rate,
-        framework='xtts',
-        language='de',
-        speaker=german_speaker,
-        return_lc3=True
-    )
-    print(f"Generated LC3 data length: {len(lc3_xtts)} bytes")
-
     print('Done.')